SkillCraft Technology Machine Learning Internship - Task 4

🖐️ Hand Gesture Recognition Using Convolutional Neural Networks (CNN)

📌 Project Overview

This project develops a Deep Learning computer vision model using a Convolutional Neural Network (CNN) to recognize and classify distinct hand gestures. Built using TensorFlow and Keras, the system processes RGB images of hand gestures, tracks spatial feature patterns, and classifies them across dynamic action categories.

📊 Dataset Structure

The system is built and trained using the Hand Gestures Dataset on Kaggle.

• Format: Multi-class RGB (color) images.
• Organization: Managed locally in an images/ directory split into unique gesture class subfolders.

⚠️ Note: The local image dataset directory is excluded from remote version tracking via .gitignore to comply with standard software hosting size constraints.

⚙️ Deep Learning Pipeline & Architecture

1. Data Preprocessing & Augmentation: Automatically scans categorical directories with opencv-python (cv2), standardizes all inputs to ((64 \times 64)) resolution, scales feature values to a normalized ([0.0, 1.0]) floating-point range, and applies one-hot encoding via Keras utilities.
2. Train-Test Split: Implements an 80/20 train-test validation partition strategy using train_test_split.
3. CNN Network Architecture Sequential Design:
   • Input Layer: Accepts ((64, 64, 3)) shape matrices.
   • Convolutional Layer 1: 32 filters, ((3 \times 3)) kernel size, ReLU activation.
   • Max Pooling Layer 1: ((2 \times 2)) pooling window downsampling spatial dimensions.
   • Convolutional Layer 2: 64 filters, ((3 \times 3)) kernel size, ReLU activation.
   • Max Pooling Layer 2: ((2 \times 2)) pooling window downsampling spatial dimensions.
   • Flattening Layer: Reshapes 2D matrices into 1D feature arrays.
   • Dense Layer (Hidden): 128 structural units using ReLU activation.
   • Regularization (Dropout): 50% neural drop probability constraint preventing overfitting.
   • Output Layer: Dynamically calculates soft-max categorical distribution metrics matched directly to individual gesture class counts.
4. Optimization Pipeline: Uses the Adam optimizer, tracking performance via categorical_crossentropy losses over 10 training epochs.
5. Analytics Reporting: Generates distinct matplotlib visualization figures plotting performance trajectories for validation accuracy and training loss.

🛠️ Local Installation & Setup

1. Clone the Repository

git clone https://github.com
cd your-repository-name

2. Install Required Dependencies

Ensure you have Python installed, then run the environment mapping:

pip install numpy opencv-python tensorflow scikit-learn matplotlib notebook

3. Dataset Configuration

1. Download the archive bundle directly from the Kaggle Hand Gestures Dataset Page.
2. Unzip the target file array and structure your project directory locally as follows:

   📁 your-repository-name
   ├── 📁 images
   │   ├── 📁 1-palm
   │   ├── 📁 2-L
   │   └── 📁 ... (all gesture folders)
   └── 📄 Task4.ipynb
   

4. Execution

Launch your local Jupyter interface to run the script:

jupyter notebook Task4.ipynb